Policy
160d agoAct-Based Approval-Directed Agents: A Policy Primer
In the 2010s, Paul Christiano pioneered the idea of act‑based approval‑directed agents, aiming to align AGIs with human intent. His Iterated Amplification framework promised a practical route to building agents that obey supervisors without deception.
The Signal
The proposal remains controversial, sparking debate over policy and safety implications in the broader AI governance context.